Really enjoyed “Are we running too many agents” hosted by @exedev, with great talks from Josh Synder from Josh Wills from Datology AI, and @beyang from @AmpCode. A major topic was about the human side of working with agents (sometimes with too many agents lol) I resonate with the main point by Josh from - running more agents usually means more context switching and giving more attention to them instead of focusing more on understanding your work. We definitely need to understand the agent systems we’re using with more attention, and take full responsibility for what it does. Also learnt a lot from Josh Wills’ advice: use agents to complement our bad habits at work. For example, we can ask for simplicity from our agents when we often overcomplicate tasks. Then @beyang’s talk about Amp shares some interesting finding about the agent workflow: making parallel work effective means improving the whole agent setup, including managed sandboxes and better ways for users to interact with Agents workflow. Agents could generate multiple options so users can compare at once, and they can show diagrams to help us understand the changes. These are all parts that make human–agent collaboration more useful in practice. Excited to explore how StepFun can support workflows like these.